Almost any IT department worth its salt is deploying virtualisation technology to reduce power usage, make server and OS deployments more flexible, and better use storage and systems resources. But as virtualisation technology gains in popularity, it may bring with it new risks, says Don Simard, commercial solutions director at the US National Security Agency. At the same time, virtualisation technology may bring new protections, he says.
One of the NSA's roles is to work with technology providers to help them make their wares more secure, both to help government agencies using them and to reduce threats that could affect the US commercial sector and thus the nation's economy. Sometimes, the NSA also wants to ensure it has back-door access to commercial systems.
In the case of virtualisation, the NSA has worked with EMC's VMware unit, IBM, AMD, Trusted Computing Group, and others for several years to identify potential threats and suggest workarounds. Later this year, chips from AMD and Intel will include technology that the NSA has helped develop.
The hidden hardware threat
Simard is a big fan of virtualisation. The technology has helped NSA employees, as well as other military and intelligence agents, access multiple secure networks from a single computer. It used to be that each network had to be accessed from a separate computer — the PC or laptop essentially acted as a hardware authentication token — so analysts and coordinators had to move from one computer to another, depending on which intelligence network they were using at the time. This led to equipment shortages and lots of boxes to carry around when travelling. In Simard's case, that meant using four computers, one each for the three intelligence networks he works on and one for unsecured, personal internet access. Now he has one computer, with each network accessed from a separate virtual machine.
But the NSA realised that this benefit of virtualisation also introduced a new potential threat. After all, Simard says, "graphics cards and network cards today are really miniature computers that see everything in all the VMs." In other words, they could be used as spies across all the VMs, letting a single PC spy on multiple networks. Although he's not aware of any such spyware today, it's not a problem the NSA wants to experience or see happen in other intelligence agencies.
That's where IBM and AMD come in. AMD's scientists had similar concerns to the NSA's, so they worked with the NSA to design an authentication mechanism at the chip level that would be able to control what hardware could do with the virtualisation engines that rely on their AMD-V on-chip virtualisation assistance technology. While no shipping date has been announced, a new generation of AMD-V chips expected later this year will introduce the concept of chip-managed trusted hardware, says Steve McDowell, division manager for emerging technologies at AMD. Intel is expected to ship a similar technology as well, says Kurt Roemer, chief security strategist at Citrix, which recently bought hypervisor maker Xen.
These new chips will have what AMD's McDowell calls a "device exclusion vector" that can authorise or block hardware access to VMs, as well as create a chain of permissions that flow from one device to another, so OS and hypervisor developers can control not only what hardware can do what, but also what flows among hardware devices are permitted. McDowell expects this approach to prevent the subsystem-as-spy problem that both it and the NSA identified.
Using virtual layers to add security
While virtualisation is used commercially to have multiple operating systems run on one machine — to get more usage from physical servers, to run Windows on Macs, and to easily set up testbed environments — its origins trace back to a military security need. In fact, the VMware technology that popularised virtualisation is a spin-off of US Defense Department-sponsored research done at Stanford University; the military saw early promise in virtual machines to encapsulate networks and desktops from outside threats, resulting in an NSA-created OS called NetTop that in 2001 did for Linux what products such as Parallels Desktop and VMware Desktop do today: provide separate VMs that can't affect each other on one box.
Now the NSA sees virtualisation protecting systems in a new layering approach, Simard says. The idea is to have an independent layer handle security, so even if an OS has security flaws, a separate layer that the OS can't compromise handles security threats such as viruses and worms or implements firewalls. Simard says it's inevitable that PC operating systems will have security holes: "The PC platform is a very feature-rich platform, and being feature-rich gets it into trouble."
The NSA, working with General Dynamics and IBM, has developed the first version of this technology, which it calls the High Assurance Platform workstation, for the US Special Operations Command, using VMware, Novell SuSE Linux, and Red Hat Linux, Simard says.
"I believe strongly in doing antivirus and firewalling in isolation outside the OS," says AMD's McDowell. But Simard is concerned that this layered approach could compromise security if poorly implemented in commercial systems. The reason: If the security layer is compromised, such as through poor design, then an intruder now has access to all the VMs on the system. McDowell agreed with that concern, saying that such a layered approach can't replace security at the OS and network — instead, it must supplement those components' security. He also notes that applications are the most common route for vulnerabilities to find their way into an OS, so they too need to have their own protection mechanisms.
A related concern is the hypervisor, the root layer that manages the VMs. If compromised, it could expose everything on the system. But McDowell is least worried about this scenario: "Hypervisors are very hard to write, and there are just three of them — Xen, Microsoft, and VMware" — so there's not broad expertise for hackers to tap into, he said.
Citrix's Roemer notes that the NSA's risk examples are on the extreme side. "They're on to something there, but a lot of their needs greatly exceed that of other organisations", he says.
The NSA's Simard agrees, but notes that there's a leapfrog effect, in which the NSA and other government agencies sometimes are the first to come across a threat, and feed that experience to commercial companies to help them improve their products. The commercial companies take the issue a step further and end up having better options than the government, which then pushes the envelope in its usage and discovers new issues.
He sees this being very true in the virtualisation world, where the Federal Government was the first to see the technology as a security aid and then, more recently, as a new potential threat vector. "Hopefully, industry will learn from our worries," Simard says.